Understanding the COVID-19 Data Quality Problem with Sherri Rose - #374
Topics covered
Popular Clips
Episode Highlights
Fairness Issues
Sherri Rose discusses the critical issue of fairness in healthcare algorithms, particularly how risk adjustment formulas can inadvertently discriminate against marginalized groups. She explains that these formulas, which are designed to distribute healthcare funds, often lead to undercompensation for groups like those with mental health and substance use disorders. This undercompensation incentivizes insurers to discriminate against these groups by altering provider availability or increasing copays 1. Rose emphasizes the importance of creating fairer formulas to reduce such discrimination, stating,
If we could redistribute the funds within the formula such that individuals with mental health and substance use disorders were not massively undercompensated, then the insurers would have less of an incentive to try and change their plans to harm those enrollees.
---
She is optimistic about the potential impact of these advancements on global health and social policy 2.
Algorithmic Fairness
Rose highlights the significant improvements in group fairness achieved through her research, noting that fairness for certain groups can be enhanced by over 98% with minimal loss in global model performance 3. She collaborates with colleagues to address fairness across multiple groups, ensuring that improvements for one group do not negatively impact others. In discussing algorithmic fairness, Rose criticizes the current state of clinical practice, where fairness is often overlooked in published papers 4. She stresses the need for multiple metrics to assess algorithms, particularly their impact on marginalized groups, and notes,
The concept of studying algorithms for issues of fairness is starting to make a dent. But when you look at published papers, the vast, vast, vast majority of published papers in clinical journals do not even consider it.
---
Rose is actively working to drive the community towards considering these metrics in healthcare research.